IMPACT OF HYPERPARAMETERS ON CNN PERFORMANCE FOR SHORT-TERM ELECTRICITY LOAD FORECASTING
DOI:
https://doi.org/10.62985/j.huit_ojs.vol26.no2E.391Từ khóa:
Short-term electricity load forecasting, 1D convolutional neural network, Hyperparameter tuning, Time series forecastingTóm tắt
This paper investigates the effects of key hyperparameters on the forecasting accuracy of a one-dimensional convolutional neural network for short-term electricity load forecasting. A univariate load time series is first transformed into supervised learning samples using a 24-step sliding window. From these generated samples, the most recent 840 samples are retained to reflect recent operating patterns. These 840 samples are then split chronologically into 672 training samples (80%) and 168 testing samples (20%). On the training set, three-fold Time-Series cross-validation is employed to preserve temporal order. A total of 160 configurations are examined by varying the learning rate, number of convolutional filters, kernel size, and dropout rate, while keeping the number of dense units, batch size, and training epochs fixed. Forecasting performance is measured using mean absolute percentage error (MAPE). The results indicate that hyperparameter choices significantly affect both accuracy and fold-to-fold stability; the best configuration achieves a mean cross-validated MAPE of approximately 2.13% with a standard deviation of about 0.44%. These findings underscore the importance of meticulous hyperparameter selection in enhancing the reliability of compact 1D convolutional models for short-term load forecasting.
Tài liệu tham khảo
[1] N. Shabbir, R. Ahmadiahangar, A. Rosin, M. Jawad, J. Kilter, and J. Martins, “XGBoost-based short-term electrical load forecasting considering trends and periodicity in historical data,” in 2023 IEEE Int. Conf. Energy Technol. Future Grids (ETFG), 2023, pp. 1–6, doi: https://doi.org/10.1109/ETFG55873.2023.10407926
[2] U. I. Lezama Lope, A. Benavides-Vázquez, G. Santamaría-Bonfil, and R. Z. Ríos-Mercado, “Fast and efficient very short-term load forecasting using analogue and moving average tools,” IEEE Lat. Am. Trans., vol. 21, no. 9, pp. 1015–1021, 2023, doi: https://doi.org/10.1109/TLA.2023.10251808.
[3] M. Pleños, “Time series forecasting using Holt-Winters exponential smoothing: Application to abaca fiber data,” Zesz. Nauk. SGGW Warszawie - Probl. Rol. Światowego., vol. 22, no. 2, pp. 17–29, 2022, doi: https://doi.org/10.22630/PRS.2022.22.2.6.
[4] U. Samal and A. Kumar, “Enhancing software reliability forecasting through a hybrid ARIMA-ANN model,” Arab. J. Sci. Eng., vol. 49, pp. 7571–7584, 2024, doi: https://doi.org/10.1007/s13369-023-08486-1.
[5] N. T. N. Anh, N. N. Anh, T. N. Thang, V. K. Solanki, R. G. Crespo, and N. Q. Dat, “Online SARIMA applied for short-term electricity load forecasting,” Appl. Intell., vol. 54, no. 1, pp. 1003–1019, 2024, doi: https://doi.org/10.1007/s10489-023-05230-y.
[6] V. Gayathry, D. Kaliyaperumal, and S. R. Salkuti, “Seasonal solar irradiance forecasting using artificial intelligence techniques with uncertainty analysis,” Sci. Rep., vol. 14, no. 1, pp. 1–19, 2024, doi: https://doi.org/10.1038/s41598-024-68531-3.
[7] M. Safari, A. H. Rabiee, and J. Joudaki, “Developing a support vector regression model for prediction of main and lateral bending angles in laser tube bending process,” Materials., vol. 16, no. 8, 2023, doi: https://doi.org/10.3390/ma16083251.
[8] A. Ajder, H. A. A. Hamza, and R. Ayaz, “Wavelet-enhanced hybrid LSTM-XGBoost model for predicting time series containing unpredictable events,” IEEE Access., vol. 13, no. April, pp. 58671–58679, 2025, doi: https://doi.org/10.1109/ACCESS.2025.3556540.
[9] M. Goyal and M. Pandey, “A systematic analysis for energy performance predictions in residential buildings using ensemble learning,” Arab. J. Sci. Eng., 2020, doi: https://doi.org/10.1007/s13369-020-05069-2.
[10] L. Semmelmann, S. Henni, and C. Weinhardt, “Load forecasting for energy communities: A novel LSTM-XGBoost hybrid model based on smart meter data,” Energy Inform., vol. 5, no. 1, pp. 1–21, 2022, doi: https://doi.org/10.1186/s42162-022-00212-9.
[11] L. Zhang and D. Jánošík, “Enhanced short-term load forecasting with hybrid machine learning models: CatBoost and XGBoost approaches,” Expert Syst. Appl., vol. 241, 2024, doi: https://doi.org/10.1016/j.eswa.2023.122686.
[12] X. Yao, X. Fu, and C. Zong, “Short-term load forecasting method based on feature preference strategy and LightGBM-XGBoost,” IEEE Access., vol. 10, no. June, pp. 75257–75268, 2022, doi: https://doi.org/10.1109/ACCESS.2022.3192011.
[13] A. Faustine, N. J. Nunes, and L. Pereira, “Efficiency through simplicity: MLP-based approach for net-load forecasting with uncertainty estimates in low-voltage distribution networks,” IEEE Trans. Power Syst., vol. 40, no. 1, pp. 46–56, 2024, doi: https://doi.org/10.1109/TPWRS.2024.3400123.
[14] A. O. Aseeri, “Effective RNN-based forecasting methodology design for improving short-term power load forecasts: Application to large-scale power-grid time series,” J. Comput. Sci., vol. 68, no. February, p. 101984, 2023, doi: https://doi.org/10.1016/j.jocs.2023.101984.
[15] K. Zhu, Y. Li, W. Mao, F. Li, and J. Yan, “LSTM enhanced by dual-attention-based encoder-decoder for daily peak load forecasting,” Electr. Power Syst. Res., vol. 208, no. February, p. 107860, 2022, doi: https://doi.org/10.1016/j.epsr.2022.107860.
[16] Z. Liang, R. Mieth, and Y. Dvorkin, “Operation-adversarial scenario generation,” Electr. Power Syst. Res., vol. 212, no. October 2021, p. 108451, 2022, doi: https://doi.org/10.1016/j.epsr.2022.108451.
[17] C. Wang, X. Li, Y. Shi, W. Jiang, Q. Song, and X. Li, “Load forecasting method based on CNN and extended LSTM,” Energy Rep., vol. 12, no. August, pp. 2452–2461, 2024, doi: https://doi.org/10.1016/j.egyr.2024.07.030.
[18] T. H. Bao Huy, D. N. Vo, K. P. Nguyen, V. Q. Huynh, M. Q. Huynh, and K. H. Truong, “Short-term load forecasting in power system using CNN-LSTM neural network,” in 2023 IEEE Asia Meet. Environ. Electr. Eng. (EEE-AM), 2023, pp. 1–6, doi: https://doi.org/10.1109/EEE-AM58328.2023.10395221.
[19] L. Li, X. Su, X. Bi, Y. Lu, and X. Sun, “A novel Transformer-based network forecasting method for building cooling loads,” Energy Build., vol. 296, no. July, p. 113409, 2023, doi: https://doi.org/10.1016/j.enbuild.2023.113409.
[20] Z. Tian, W. Liu, W. Jiang, and C. Wu, “CNNs-Transformer based day-ahead probabilistic load forecasting for weekends with limited data availability,” Energy., vol. 293, no. February, p. 130666, 2024, doi: https://doi.org/10.1016/j.energy.2024.130666.
[21] T. T. Ngoc, L. Van Dai, and C. M. Thuyen, “Support vector regression based on grid search method of hyperparameters for load forecasting,” Acta Polytech. Hung., vol. 18, no. 2, pp. 143–158, 2021, doi: https://doi.org/10.12700/APH.18.2.2021.2.8.
[22] T. N. Tran, “A new grid search algorithm based on median values for SVR model in case of load forecasting,” Period. Polytech. Electr. Eng. Comput. Sci., vol. 67, no. 1, pp. 51–60, 2023, doi: https://doi.org/10.3311/PPee.20887.
